Waldkirch: Thesis: Image-based object segmentation using AI accelerators on embedded camera systems

Winter semester 2026/27 – fixed term of 3–6 months

Precise assignment of detected features (e.g., barcodes) to the correct objects (e.g., parcels) is essential in logistics. Currently, only the positions of the features within the image are known – not the positions of the actual objects. This makes reliable assignment challenging, especially for objects located close to each other. To improve assignment accuracy, the object region in the image should be identified. By correlating the detected object region with the feature position, the reliability of the assignment can be significantly increased. The goal of this project is to integrate a demonstrator on an embedded camera system that can segment and localize individual objects in real time based on image sequences of material flows.

YOUR TASKS:

• Commission a KI accelerator on an embedded camera system
• Implement AI-based segmentation for object detection using the KI accelerator
• Deploy a fully functional solution on the embedded camera system
• Evaluate the developed solution in a real application environment

YOUR PROFILE:

• Ongoing studies in Computer Science, Computer Vision, Machine Learning, or a comparable field
• Basic knowledge of optics or physics
• Programming experience in Python, MATLAB, Lua, or C++
• Independent and structured working style, with a quick grasp of complex topics
• Strong teamwork and communication skills

Contact: Sarah Disch